This library implements fully vectorized Beam Search, Greedy Search and sampling for sequence models written in PyTorch. This is specially useful for tasks in Natural Language Processing, but can also be used for anything that requires generating a sequence from a sequence model.
from pytorch_beam_search import autoregressive
# Create vocabulary and examples
# Tokenize the way you need
corpus = list("abcdefghijklmnopqrstwxyz ")
# len(corpus) == 25
# An Index object represents a mapping from the vocabulary
# to integers (indices) to feed into the models
index = autoregressive.Index(corpus)
n_gram_size = 17 # 16 with an offset of 1
n_grams = [corpus[i:n_gram_size + i] for i in range(len(corpus))[:-n_gram_size + 1]]
# Create tensor
X = index.text2tensor(n_grams)
# X.shape == (n_examples, len_examples) == (25 - 17 + 1 = 9, 17)
# Create and train the model
model = autoregressive.TransformerEncoder(index) # just a PyTorch model
model.fit(X) # basic method included
# Generate new predictions
new_examples = [list("new first"), list("new second")]
X_new = index.text2tensor(new_examples)
loss, error_rate = model.evaluate(X_new) # basic method included
predictions, log_probabilities = autoregressive.beam_search(model, X_new)
# every element in predictions is the list of candidates for each example
output = [index.tensor2text(p) for p in predictions]
outputfrom pytorch_beam_search import seq2seq
# Create vocabularies
# Tokenize the way you need
source = [list("abcdefghijkl"), list("mnopqrstwxyz")]
target = [list("ABCDEFGHIJKL"), list("MNOPQRSTWXYZ")]
# An Index object represents a mapping from the vocabulary
# to integers (indices) to feed into the models
source_index = seq2seq.Index(source)
target_index = seq2seq.Index(target)
# Create tensors
X = source_index.text2tensor(source)
Y = target_index.text2tensor(target)
# X.shape == (n_source_examples, len_source_examples) == (2, 11)
# Y.shape == (n_target_examples, len_target_examples) == (2, 12)
# Create and train the model
model = seq2seq.Transformer(source_index, target_index) # just a PyTorch model
model.fit(X, Y, epochs=100) # basic method included
# Generate new predictions
new_source = [list("new first in"), list("new second in")]
new_target = [list("new first out"), list("new second out")]
X_new = source_index.text2tensor(new_source)
Y_new = target_index.text2tensor(new_target)
loss, error_rate = model.evaluate(X_new, Y_new) # basic method included
predictions, log_probabilities = seq2seq.beam_search(model, X_new)
output = [target_index.tensor2text(p) for p in predictions]
output- The greedy_search function implements Greedy Search, which simply picks the most likely token at every step. This is the fastest and simplest algorithm, but can work well if the model is properly trained.
- The sample function implements sampling from a sequence model, using the learned distribution at every step to build the output token by token. This is very useful to inspect what the model learned.
- The beam_search function implements Beam Search, a form of pruned Breadth-First Search that expands a fixed number of the best candidates at every step. This is the slowest algorithm, but usually outperforms Greedy Search.
- The autoregressive module implements the search algorithms and some architectures for unsupervised models that
learn to predict the next token in a sequence.
- LSTM is a simple baseline/sanity check.
- TransformerEncoder is a GPT -like model for state-of-the-art performance.
- The seq2seq module implements the search algorithms and some architectures for supervised encoder-decoder models
that learn how to map sequences to sequences.
- LSTM is a sequence-to-sequence unidirectional LSTM model similar to the one in Cho et al., 2014, useful as a simple baseline/sanity check.
- ReversingLSTM is a sequence-to-sequence unidirectional LSTM model that reverses the order of the tokens in the input, similar to the one in Sutskever et al., 2014. A bit more complex than LSTM but gives better performance.
- Transformer is a standard Transformer model for state-of-the-art performance.
pip install pytorch_beam_searchThe project is licensed under the MIT License.